I've done CL liquid ultra about 8 times on different things and I always "wet" both surfaces before putting them together. It flows and wets onto copper like you're soldering. I've never had to heat the surfaces with the CL stuff. It's weird how it actually wets to the bare dies too. I've done GPUs, delids, and IHS to waterblock. All have been in use for 6-8+ years with no known problems.

edit On some IHS to waterblock applications, I have block sanded the plating off the IHS to apply CL directly to the copper. All of the waterblocks were copper, as were the GPU heatsinks

 

Has anyone had any success pre-forming a smooth gallium/copper layer on a fresh copper plate prior to installing a cooler?

Yes, but it takes forever, to the extent that I don't bother with liquid metal on pure copper anymore unless I have a couple of weeks to treat the part involved.

When I do have the time, I apply a generous layer of LM to the surface, let it sit somewhere warm for a week, wipe off any excess liquid (usually isn't any the first go), buff smooth any crystallization that might cause unevenness (usually with bronze wool), then apply another layer and come back next week. After three or four repeats of this I then assemble the part, using about twice the LM I would with nickel plated components (I usually have everything in the vicinity well protected)...then it ill generally last forever (or at least years).

I don't think I'd bother with pure gallium, under the assumption that it could dilute the components of the final LM TIM, which could, in turn, reduce the usable temperature range of the TIM, causing it to completely solidify when cold, which could then result in fractures, air intrusions, and voids. Then again, since it's primarily the gallium that alloys with the copper, maybe not using pure gallium skews the composition even more. Regardless, I typically use use the liquid metal TIM itself, because I have a lot of cheap Phobya stuff laying around.

 

Yes, but it takes forever, to the extent that I don't bother with liquid metal on pure copper anymore unless I have a couple of weeks to treat the part involved.

When I do have the time, I apply a generous layer of LM to the surface, let it sit somewhere warm for a week, wipe off any excess liquid (usually isn't any the first go), buff smooth any crystallization that might cause unevenness (usually with bronze wool), then apply another layer and come back next week. After three or four repeats of this I then assemble the part, using about twice the LM I would with nickel plated components (I usually have everything in the vicinity well protected)...then it ill generally last forever (or at least years).

That's useful info - as I suspected, it's something that works but takes a looong time to implement. I have a spare NH-U14S I could make do with while 'treating' the NH-D15S, but it still sounds like a lot of patience needed. Then again, I have no intention of upgrading from the 9700K for a few years, and if the results proved rewarding enough it might be worth it (it's quietitude I'm after, not dizzying overclocks - if I can keep the three 14cm fans involved under 1000RPM while the CPU is pulling 130-140 watts video encoding I'll be happy.

ETA >> I wonder if high temperatures (say. c. 70-80°C and using gloves) and lots of agitation would speed the reaction up? Hmmm, I'm still tempted to try, only perhaps on a piece of copper sheet before I grind the nickel off the Noctua.

I don't think I'd bother with pure gallium, under the assumption that it could dilute the components of the final LM TIM, which could, in turn, reduce the usable temperature range of the TIM, causing it to completely solidify when cold, which could then result in fractures, air intrusions, and voids. Then again, since it's primarily the gallium that alloys with the copper, maybe not using pure gallium skews the composition even more. Regardless, I typically use use the liquid metal TIM itself, because I have a lot of cheap Phobya stuff laying around.

Indeed. But possibly the actual mass of gallium finally absorbed during 'treatment' to form the 'barrier layer' might be proportionally small enough to the amount of LM applied that further migration of the other metals might not be a problem (wish I knew the proportions in LM).

 

There is a way to speed the curing process up substantially.

But you have to make sure that the cold plate isn't attached to any plastics or electronics (e.g if you can remove the cold plate completely, or it's only attached to some sort of heatpipe but nothing more. This works BEST with IHS lids (perfect, in fact), as no extra work is needed. Other than that, you would have to somehow completely remove the cold plate and only have the copper available, with no other parts.

What you're supposed to do is apply LM to the cold plate, spread it around completely, then bake it in an oven at 110C for about 2-3 hours.
Then let it cool enough to handle, then remove the liquid metal from the cold plate, leaving the silver stain behind on it. Then reattach the cold plate to the cooler, apply a new coat of liquid metal, and the LM should last for years.

If it isn't possible to bake the LM, then the second best option is to wipe the contact surface (if there is a 2nd copper surface, wipe that too) with 1500 grit sandpaper, using your fingers, to roughen up the entire surface with micro-grooves.

Then, clean it completely and fully, then apply a coat of liquid metal on it, then spend about 15 (!) minutes wiping the liquid metal completely around the plate, over and over, until the consistency begins to change into a loose paste like consistency, rather than a liquid. (basically, when you can hold the surface vertically and no LM moves anywhere). Remember you want to use only a little.
Then once you finish, apply another layer of LM (do not clean off the first one, unlike the baking method!) on top of the first, spread it around, apply a matching layer to the other surface (always do this), then assemble the cooler.

 

e.g if you can remove the cold plate completely, or it's only attached to some sort of heatpipe but nothing more.

I've turned flat heatpipes permanently back into round ones by heating them too much.

110C should be safe though.

but how far in does one need to go before it just becomes an academic or scientific exercise?

Modern CPU dies are approaching thermal densities of two watts per square mm. This is like Bunsen burner levels of thermal flux.

Too much (reduction of thermal resistance) will never be enough for some.

 

Is there seriously a benefit to adding LM to a processor in this manner?

I know lapping a processor has some tangible benefit, I've done it myself in years long passed and it did a lot of good... but how far in does one need to go before it just becomes an academic or scientific exercise?

 

Is there seriously a benefit to adding LM to a processor in this manner?

I know lapping a processor has some tangible benefit, I've done it myself in years long passed and it did a lot of good... but how far in does one need to go before it just becomes an academic or scientific exercise?

It's substantial. Both of my 9900K's are direct die with LM. It is really hard to do this generation though with how tiny the actual size of the die is. I don't know if there are any other solutions besides the suprercool kit.

 

Makes one wonder if there'd be a market for coolers with pre-'galliumised' contact plates.

 

I broke my 2.8 Ghz AMD Athlon-M Barton 2600+ (IQYHA0401UPMW) the other day because I forgot to pre-heat the CPU/Heatsink before trying to remove it.

The Thermalright copper block had "soaked in" all the LM (Conductonaut) and when I went to remove it it broke a piece of silicon from the CPU that it had bonded to

A careless mistake, now am on the hunt for a replacement for my collection (damn dispise those German ebay sellers who cannot be bothered to answer back to questions regards the products they are selling).

Still have a 2800+ but its not a mobile, so cant push it anywhere near 2.8 Ghz, max I got out of it was 2.6 something.

Just something to be aware of with regards to LM and bonding to silicon ...

 

I may have a Barton core 3000+ if interested I just have to check the collection.

what kind of performance differences are u guys seeing with this mod?